Optimise AI

AI Optimisation

Better answers, lower cost per query, fewer escalations to a human.

A four-week engagement to improve the accuracy, cost and reliability of AI already running in your organisation, measured against an agreed baseline.

Duration

4 weeks

Pricing

on request

Deliverables

8 included

The figure above covers this initial engagement. Where discovery shows the work needs more, we continue on an optional time-and-materials basis.

What's included

  • Baseline evaluation of accuracy, cost and latency
  • Failure case collection and error analysis
  • Prompt, retrieval and context tuning
  • Model selection and routing review
  • Token cost and latency optimisation
  • Guardrail and escalation path hardening
  • Monitoring and reporting for the system owner
  • Before-and-after measurement against the agreed baseline

Who it's for

  • AI deployments that work but cost more than anyone expected
  • Assistants giving inconsistent answers users have stopped trusting
  • Teams with no measurable view of whether their AI is getting better or worse

AI systems degrade quietly. Prompts accumulate patches for problems that no longer exist, retrieval pulls in documents that were reorganised months ago, and the model everyone chose at launch is now neither the cheapest nor the best for the job. Meanwhile the cost per query drifts upward and nobody owns the number.

This is a focused engagement on an AI system you already run — no replatforming, no new use cases.

How it runs

Week 1 — Baseline. We measure what you have: accuracy against a representative test set, cost per query, latency, and escalation rate. We collect real failure cases from actual usage, because the useful errors are rarely the ones anyone thought to test for.

Weeks 2–3 — Tune. We work the levers in order of payback — prompt and context structure, retrieval quality, model selection and routing, then guardrails and escalation paths. Each change is measured against the Week 1 baseline rather than accepted on impression.

Week 4 — Measure and hand over. We re-run the full evaluation, report the before-and-after, and hand over monitoring and documentation your team can maintain.

We measure before we change anything

Without a baseline, “improving” an AI system is guesswork, and it is entirely possible to make it worse while feeling confident. The Week 1 measurement is what makes the rest of the engagement defensible — and it is what lets you say something specific to your board about what changed.

What we optimise for

We agree the target metrics at the start. Usually that is answer accuracy, cost per query, and escalation rate — but if your pressure is latency, or a specific class of failure that is eroding user trust, we measure that instead.

Where it fits

If the system is not in production yet, AI Pilot to Production is the engagement you want. If you are not sure which of several AI initiatives deserves the investment, start with the AI Readiness Assessment. Wider context on our approach is under AI Consulting and Agentic AI.

Interested in AI Optimisation?

Tell us about your environment and we will confirm the starting scope and price before any commitment. If a different package fits better, or if you need something else entirely, we will say so.